| Summary: | This thesis presents advanced control strategies to improve the dynamic performance and efficiency of Silicon Carbide (SiC)-based Permanent Magnet Synchronous Motor (PMSM) drive systems, specifically for high-performance electric racing applications. It begins with the development of a comprehensive analytical loss model that captures the nonlinear behaviour of both the machine and power electronics. This model offers detailed insights into energy dissipation and serves as the foundation for optimising $dq$-axis current references. Experimental validation shows that predicted system losses align with measurements within a Normalized Root Mean Square Error (NRMSE) below 5\%, confirming its suitability for efficiency optimisation.
Building on this, an online polynomial fitting-based Loss Minimisation (LM) strategy is introduced. This method enables real-time adjustment of the $dq$-axis current without relying on precomputed lookup tables, significantly improves flexibility. It maintains a tracking error within 0.5\% under rated load and reaches the optimal operating point within a single control cycle, whereas conventional online methods may take up to 1500 cycles to converge.
The LM strategy is then integrated with Finite Control Set Model Predictive Control (FCS-MPC), resulting in notable gains in both dynamic response and energy efficiency. By reducing switching losses, it achieves up to 1.1kW loss reduction at rated load. To enhance robustness, an online parameter estimation method using virtual small signal injection is proposed. This method identifies key parameters—inductance, resistance, and flux linkage—with just 82 floating-point operations per cycle, ensuring fast, real-time adaptation and mitigating the effects of parameter mismatch.
Finally, the LM strategy is embedded in an enhanced Model Predictive Torque Control (MPTC) framework, unifying torque and loss control in a single loop. Experimental results show tracking accuracy within 2\% and a dramatic reduction in dynamic settling time—from over 10ms in conventional online LM methods to just 0.17ms with the proposed strategy—demonstrating a substantial improvement in transient performance.
|
|---|